ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Vectorized Transpose for Batched Transpose CK Tile Operator (#2131)

Browse Source 
* Shared Memory for single data point

* CKTile Transpose vectorize CP1

* CKTile Transpose vectorize CP2

* CKTile Transpose vectorize CP2.1

* fixed the compile error of the transpose tile 2d

* Have the correct result for the current test sample

* Changes to printing tensor

* fp8 support added

* Debugging for transpose

* solving the corner issue

* Changed padding flag

* Intermideate Debugging

* Intermidiate Debugging

* Intermediate Debugging

* Finished debugging of the transpose op

* Code Cleanup

* Adding edge case smoke tests

* Adding Transpose test to CI/CD

* Adding Transpose test to CI/CD

* Adding Transpose test to CI/CD

* Addressing Review Comment

* Addressing Comments

* Addressing Comments

* Measuring Perf Tests

* Code Cleanup

* Changlog

* Added the running iterations

* clang format

* Fix the changelog

* Fix the compilation error

* change the printing factor

---------

Co-authored-by: ThruptiRajLakshmanaGowda <tlakshma@amd.com>
This commit is contained in:
Thomas Ning
2025-05-12 00:41:45 -07:00
committed by GitHub
parent d8faf1c6a1
commit 9d1e44e56a
14 changed files with 311 additions and 152 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
example/ck_tile/35_batched_transpose/script/perf_test.sh Executable file
View File

@@ -0,0 +1,11 @@
#!/bin/sh
EXE=./build/bin/tile_example_batched_transpose
for pr in "fp8" "fp16" "bf16"; do
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=1024 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=1024 -H=1 -W=2048 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=4096 -H=1 -W=2048 -layout_in='NCHW' -layout_out='NHWC'
done

38
example/ck_tile/35_batched_transpose/script/run_full_test.sh Executable file
View File

@@ -0,0 +1,38 @@
#!/bin/bash
#
# in order to run this script you'd first need to build the tile_example_batched_transpose executables in ../build/bin/
#
# run the script as "./run_full_test.sh <tag for your test environment> <branch name> <host name> <gpu_arch>
# input arguments:
# environment tag : a string describing the specifics of your test environment
# branch name : name of the branch in git repo (git status | grep -e 'On branch')
# host name : $hostname
# gpu architecture: e.g., gfx90a, or gfx942, etc.
#get the command line arguments:
export env_type=$1
echo 'Environment type: ' $env_type
export branch=$2
echo 'Branch name: ' $branch
export host_name=$3
echo 'Host name: ' $host_name
export GPU_arch=$4
echo 'GPU_arch: ' $GPU_arch
function print_log_header(){
rm -f $1;
echo 'On branch ' $3 &> $1;
echo 'Node name: ' $4 >> $1;
#get GPU_arch and number of compute units from rocminfo
echo -n "GPU_arch: " >> $1; rocminfo | grep "Name:" | grep "gfx" >> $1;
rocminfo | grep "Compute Unit:" >> $1;
hipcc --version | grep -e 'HIP version' >> $1;
echo 'Environment type: ' $2 >> $1;
/opt/rocm/bin/amdclang++ --version | grep -e 'InstalledDir' >> $1;
}
#run verification tests
example/ck_tile/35_batched_transpose/script/smoke_test.sh
#run performance benchmarks

20
example/ck_tile/35_batched_transpose/script/smoke_test.sh
View File

@@ -2,10 +2,26 @@
EXE=./build/bin/tile_example_batched_transpose
for pr in "fp32" "fp16" "int8" ; do
for pr in "fp8" "fp16" "bf16"; do
$EXE -pr=$pr -N=1 -C=32 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=2 -C=12 -H=1 -W=32 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=3 -C=1334 -H=1 -W=37 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=4 -C=27 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=5 -C=1234 -H=1 -W=12 -layout_in='NCHW' -layout_out='NHWC'
done
$EXE -pr=$pr -N=1 -C=1 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=1 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=128 -C=1024 -H=64 -W=64 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=128 -C=1024 -H=64 -W=64 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=16 -C=64 -H=32 -W=128 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=16 -C=64 -H=128 -W=32 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=1 -C=2048 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=2048 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=1 -C=1 -H=1024 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=1 -H=1024 -W=1024 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=8 -C=16 -H=8 -W=16 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=8 -C=16 -H=8 -W=16 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -N=1 -C=64 -H=1024 -W=1 -layout_in='NHWC' -layout_out='NCHW'
done